HISTORIA MATHEMATICA 4 (1977), 445-452

SOMECOMMENTS ON R, J, GILLINGS' ANALYSISOF

THE 2/N TABLEIN THE RHINDPAPYRUS

BY M, BRUCKHEIMER AND Y, SALOMON,

SCIENCE TEACHING CENTER, WEIZMANN INSTITUTE, ISRAEL

Summaries

This paper is a critique of R. J. Gillings,

Mathematics in the Time of the Pharoahs, Chapter
6. There are three parts: the first is devoted
to general historical comments, the second deals
with mathematical and computational points. The
third consists of two appendices, which provide
some of the detailed background for the comments
made in the body of the paper.

Cet article est une critique du sixieme

chapitre du livre de R. J. Gillings, Mathematics
in the Time of the Pharaohs. I1 se divise en
trois parties. lsns la premiere, nous y faisons
quelques commentaires historiques d'ordre g&&al.
La seconde partie est consacree h l'dtude des
aspects mathdmatiques et calculatifs. La troisieme
partie est form&e de deux appendices qui
fournissent certains renseignements sur le
cheminement pr6cis ayant men6 aux commentaires
faits dans le corps de l'article.

In Chapter 6 of Mathematics in the Time of the Pharaohs

[1972], Gillings gives an extended analysis of the table on the
Recta of the Rhind Mathematical Papyrus (RMP). For each fraction
of the form 2/n, where n is odd, the table gives an equivalent
expression as the sum of unit fractions. There are many
interesting conjectures and ideas in this analysis, but it is
unfortunately marred by occasional lapses, both historical and
mathematical. In the following we indicate some of these lapses.

I. HISTORICAL COMMENTS
If we look first at the historical aspect of the work,
Gillings seems to have fallen afoul of his own enthusiasm. His
introduction to Chapter 6 is indicative of some of what is to
follow. For example : “And how did it come about that, of all
the many thousands of possible answers to these decompositions,
those recorded by the scribe of the RMP were in almost every
Copyright @ 1977 by Academic Press, 1~.
AI[ rights of reproduction in any iOrr?lre.wwd.
446 M. Bruckheimer 6 Y. Salomon HM4

case the simplest and best possible, by his own prescribed

standards?" [p. 471
First of all, "his own prescribed standards" are the
conjectures of Gillings and other historians, and not the work
of the scribe, who makes no explicit mention of any standards.
And it is not so surprising that given a set of fifty unit
fraction expressions, out of many thousand possible ones, we
should be able to make up, post facto, some standards by which
we come to the conclusion that the scribe chose the simplest
and best possible. It is almost like making up the rules of a
game to prove that we have won, after the game is finished.
If this partiality towards the scribe were confined to
the descriptive passages, it could be forgiven as justifiable
enthusiasm for the scribe's (or his predecessors') undoubted
achievement, but it also enters the analysis. Gillings gives
five precepts "which I believe were the scribe's primary guide"
in choosing the one unit fraction decomposition for each case,
from the many available. They are as follows [p. 491:
"CANON FOR THE RECTO OF THE RMP/ PRECEPT I/ Of the possible
equalities, those with the smaller numbers are preferred, but
none as large as 1,000. [The largest number in the Recta is
890. Of the 128 numbers of the table only 11 exceed 5001.
PRECEPT 2/ An equality of only 2 terms is preferred to one of
3 terms, and one of 3 terms to one of 4 terms, but an equality
of more than 4 terms is never used. PRECEPT 31 The unit
fractions are always set down in descending order of magnitude,
that is, the smaller numbers come first, but never the same
fraction twice. PRECEPT 41 The smallness of the first number
is the main consideration, but the scribe will accept a slightly
larger first number, if it will greatly reduce the last number.
PRECEPT 5/ Even numbers are preferred to odd numbers [There
are 104 even and only 24 odd numbers used in the table], even
though they might be larger, and even though the numbers of
terms might thereby be increased." [In this extract, we have
first indications of the considerable number of minor errors in
this chapter. A simpie count verifies that there are 13, and
not 11, numbers in the table which exceed 500 and that there
are 103 even and 25 odd numbers, and not 104 and 24 as stated.
See also later].
With these precepts in mind, consider Gillings' analysis
of the case 2 f 19 [p. 581: "The only 2-term decomposition
here is 10 190 [The bar indicates the reciprocal], and the
scribe must have thought hard before he rejected it. Of the
sixteen 3-term values availage, --only five consstwholly of
even numbersA These are A 10 240 912/ B 12 48 912/
C 12 60 190/ D 12 76 ii$ E 16 24 =/. Precept 1
must have prevailed, for the scribe chose D, as having the
smallest numbers, for although 10 is less than 12, 114 is much
less than 180 (see Precept 4)."
HM 4 Comments on Gillings’ analysis 447

According to the stated precepts, there is no doubt that

the analysis is difficult, and the most that we can say is the
argument is unconvincing.
Consider another example of Gillings’ partiality for his
scribe, which will lead us on to the second part of this note,
in which we discuss the mathematical lapses in Gillings’ analvsis.
(The reference to the computer output will be taken up subse-
quently.) The extract is from pages 68-69. “2 + 95/ There
are 148 values recorded by the computer, of which 116 have 4
terms-, zhave 3Lers and 3-- have 2 terms. These last are
A 50 950/ B 57 285/ C 60 228J Now by the canon, C
should have been his immediate choice here; . . . But this is
not the value the scribe gives. So far, we have been unable
seriously to challenge the scribe’s choice of values in the
Recta Table; but here, perhaps, he faltered. “Even Homer nodded”
on occasions, The equality the scribe records is 60 380 570,
--a 3-term value which is in fact equivalent to 60 228, for
380 570 = 228, . . . . What he must have done here was to note
that 95 = 5 l 19, . . . . We cannot of course be sure that he did
not search among the 3-term decompositions having even numbers;
but if he haLhe could have found, as KDF-9 showsus,
A 56 --532 760/ B 76 152 760/ C 76 160 608/
D 76 190 380/ E 80 190 304/ and no doubt D and E would
have tempted him, by virtue of Precept 1, while Precept 4 may
have made him hesitate. We can scarcely say that the scribe
made a bad choice for 2 + 95; we can only say that he might
have expressed his answer more concisely, and that it is a great
pity he did not check with his tables.”
We feel, that this section exposes its own faults as a
piece of historical analysis, especially in the last sentence.
By the precepts laid down by Gillings, the scribe has made a
thoroughly bad choice.

II. MATHEMATICAL COMMENTS

A further reason for quoting the above section is that it
also contains a number of other errors, as we shall see a little
later. In order to carry out his analysis, Gillings had a KDF-9
computer programmed “to calculate all the possible unit-fraction
expressions of each of the divisions of 2 by the odd numbers
3, 5, 7, . . . . 101, in order to compare the decompositions given
by the scribe of the RMP with the thousands of possible forms.
Such a comparison between the calculations of an ancient
Egyptian scribe and the 22,295 values produced by a twentieth-
century computer, . . . . will undoubtedly be of great interest to
historians of mathematics .I1
Even according to Gillings’ own calculations there are
22,319 values produced by the KDF-9. But this is only a trivial
error compared with the fact that there are, in fact, many more
448 M. Bruckheimer & Y. Salomon HM4

possible unit fraction expressions--approximately 28,000.

First let us be clear what we are talking about. For any fraction
of the form 2/n we consider all the possible equivalent ex-
pressions as the sum of 2, 3, or 4 different unit fractions whose
denominators are less than 1000. This is apparently what
Gillings is talking about as well (with one minor exception on
paw 51, where he includes 45 45 among the two- term
expressions for 2 f 45).
We produced a computer program for the IBM 370-165 at the
Weizmann Institute for other work on unit fractions and we
decided to use it to check Gillings’ results. There is a
substantial difference between the results we obtained and those
given by Gillings. In Appendix A we give a table showing the
total number of unit fraction expressions obtained by Gillings
compared with those we obtained. (For an explanation of the
third set of figures, see later.) Here we shall illustrate
some of the errors that occur in Gillings’ analysis.
First of all consider the analysis of 2 5 95 which we
quo ted above. According to the results we obtained, there are
768 (not 148) values recorded by the computer, of which 720
(not 116) have 4 terms, 45 (not 29) have 3 terms, and 3 have 2
terms. Subsequently in the same analysis quoted above, Gillings
discusses 3-term decompositions having only even numbers,of
which he found five and suggests that the last two 76 190 380
and 80 190 304 may have tempted the scribe. But what about
the following: 60 300 zi/ 60 304 912/ 60 342 684/
64 228 960/ 66 228 660/ 68 228 510/ ??? 228-- z/
72 228 360/ 78 228 260/ 80 228 ?%3/ 84 210 228/
i% 180 228/ 96 160 228/ ii% = 228/ 110 132 228.
(It is possible that the last three expressions are to be
excluded because the smallest denominator exceeds that of the
original fraction; but no such rule is stated anywhere.) Had
the scribe known about these, he might have been tempted even
more.
We give one further example in detail. In discussing
2 t 29, Gillings notes the one 2-term expression and the eight
3-term expressions, none of which were chosen by the scribe.
He then states that there are 203 expressions containing 4-terms
(we could find only 197, but no matter) and continues (p. 60)
II
. ..onlQ three contain nzbers 1~ than 300. Theyare
A 24 58 i% -%?/ B 29 42 174 203/ C 29 58 87 i%/
Bv all the vrecents
I I I
of the canon he must choose A, and this is
indeed the equality recorded in the Recta.” ---
also 116 145 %/ 2014~145 203/
21 i% Butde 145 %/ f-Id3
30 42 145 z/ 30 58 87 145. (Again
the last two may need to be excluded because of the size of the
smallest denominator.)
With the above two examples (which are merely representa-
tive of many similar errors) we leave this aspect of the article,
HM4 Comments on Gillings’ analysis 449

and turn, finally, to some elementary mathematics. Gillings

devotes a short section specifically to those cases in which the
divisors are prime (p. 52-53). He remarks: “An analysis of the
computer output shows that 2-term decompositions are rare for
primes . . ..‘I and makes similar remarks in several other places.
We can improve on this, by proving very simply that if n is
prime, there is a unique 2-term decomposition for 2 5 n. The
proof is given in Appendix B. We note that if n is prime and of
the form 2k + 1, where k is a positive integer, then 2/(2k + 1) =
l/(k + 1) + l/(k + 1)(2k + 1). Thus, if we impose the condition
that a denominator shall not exceed 1000, we have
(k + 1)(2k + 1) < 1000 which gives k i 21. That is, 43 is
the last prime divisor for which a possible 2-term expression
exists (cf. Gillings, p. 53, Table 6.2). This result can be
extended quite easily to non-prime divisors (see Appendix B).
A more debatable point, not discussed by Gillings, is the
inclusion or exclusion of reducible expressions in his analysis,
where by “reducible” we mean a unit fraction expression that
can be reduced to one containing fewer terms by simple addition
of some of the terms. For example, in the case of 2 f 95,z
have, as we have seen, 60 380 570 = 60 228, and so 60 380
570 is reducible. So are all the other 3-term expressions
given by us for 2 + 95 that were omitted by Gillings. But, in
other cases, Gillings does include them; for example, for 2 + 19,
A and C are reducible. It is undoubtedly true that the inclu-
sion of reducible cases introduces an enormous number of
irrelevant expressions, since the effect of performing the
reduction is to produce a shorter expression and a smaller
denominator, and one suspects that the scribe would have discarded
many of them “at sight”. Gillings, therefore, gives the wrong
impression when he writes things like (p. 62, discussion of
2 i 43) “one can only remain lost in hopeless admiration of the
ancient Egyptian scribe, who could . . . so unerringly locate this
value among the 124 available”, when there are, in fact, only
20 irreducible expressions.
But the case is not quite clear cut, for at least three
reasons : (i) We see that the scribe did not always use irre-
ducible expressions--although he usually did; (ii) If we play
the game according to the rules proposed by Gillings then we
can replace a sisle odd denominator by two even ones;for
example 2 t 5 = 3 15 [chosen by the scribe) = 3 24 40, etc.
(iii) Reduction w produce an- inadmissable expression; for
example 2 + 29 = 29 58 87 i%= 29 58 58=z 29. so that
by the rules we should consider the 4-term expression,*because
the reduced expressions are illegal. In spite of these objections,
it is evident that the inclusion of reducible expressions intro-
duces a lot of “noise” in such a comparative study. In order to
indicate the order of magnitude of this noise we have included
a third column in the table in Appendix A, which shows the
450 M. Bruckheimer & Y. Salomon HM4

number of “irreducible” expressions. We have included among the

“irreducible” expressions those which reduce to illegal
expressions as described in (iii) above.
Finally, at the end of Chapter 6 (p. 70), Gillings cites
a paper by E. M. Bruins [1952]. We disagree with Gillings’
interpretation of this paper and suggest that it is well worth
reading for anyone interested in the subject.

APPENDIX A
Table of Total Numbers of Unit Fraction Expressions for
2/n, n = 3, 5, 7, . . . . 101.
In the following table we give the total number of 2, 3
and 4-term expressions obtained by (i) Gillings’ KDF-9 program
in column II; (ii) our IBM 370-165 program in column III;
(iii)
. . as in (ii), but after the removal of the “reducible”
expressions, in column IV.

n II III IV )ivisor II III IV

3 53 66 5 53 24 23 9
5 269 216 20 55 1128 1376 277
7 320 297 58 57 645 679 125
9 516 509 54 59 20 19 7
11 384 358 99 61 7 6 6
13 436 320 96 63 1067” 2077 284
15 1158 1179 105 65 865 1137 202
17 479 242 79 67 21 25 3
19 273 275 71 69 500 555 82
21 1190 1244 135 71 25 28 7
23 387 282 77 73 10 11 5
25 619 600 116 75 884 1390 142
27 733 768 108 77 741 1181 250
29 212 206 44 79 3 4 4
31 164 153 47 81 339 442 53
33 1018 1086 169 83 3 4 4
35 1458 1707 255 85 290 844 155
37 97 87 22 87 102 342 81
39 894 980 171 89 6 9 9
41 187 183 33 91 216 978 238
43 124 113 20 93 58 228 54
45 1967 2230 232 95 148 768 120
47 56 48 25 97 8 10 2
49 371 412 87 99 710 1442 242
51 595 778 157 101 1 1 1

* or 1607: two different values are given by Gillings (I ). .56-57)

HM4 Comments on Gillings' analysis 451

APPENDIX B
On the number of 2-term expressions for the fraction 2 z n,
where n is odd.

Suppose that 2/n = l/a + l/b. We write a = a'nl, where nl

is the HCF of a and n. Similarly b = b'n2. Then Z/n =
l/a'nl + l/b'n2 = dl/a'n + d2/b'n, where n = dlnl = d2n2, and dl
is coprime to a' and d2 is coprime to b'. Now from 2 =
al/a' + d2/b' we have 2a' = dl + d2a'/b', and since b' is coprime
to d2, we must have b' divides a'. Similarly a' divides b',
i.e. a' = b' = (dl + d2)/2.
We can therefore conclude that to every Z-term expression
for 2/n there corresponds a pair of numbers (d,, d2), where both
dl and d2 are factors of n. Conversely, to every such pair
Cdl I d2) there corresponds a 2-term expression.

(Note, since n is odd, dl + d2 is always even.) Since we

require a # b, we shall have to require d
1 + d2.
It would now seem that in order to find the number of 2-
term expressions for a given 2/n, all we have to do is count the
number of different pairs of factors (dl, d2) we can obtain from
n. But it is possible that two distinct pairs (dl, d2) and
(e lP e2) give the same expression. In fact, a little elementary
algebra shows that they will if and only if they are both multi-
ples of some other pair in the sense that (dl, d2) = d(m,, m2) =
(dml, dm2), (e,, e2) = e (ml, m2) = (eml, ema). It follows that
what we have to count is the number of different coprime pairs
of factors (dl, d2) that we can obtain from n.
Examples If p and q denote primes then we have the
following results: (i) For 2/p there is just one pair (p, 1);
1. e. the 2-term unit fraction expression is unique:

ajp;‘)+(p:,“jp
(ii) For 2/p=, where I‘ is apositive integer, there are exactly
452 M. Bruckheimer & Y. Salomon HM 4

r 2-term expressions. (iii) For 2/pq there are four 2-term

2
expressions. (iv) For 2/p q there are seven 2-term expressions.
We could obviously go on to obtain more general results
for more complicated n, but these four examples cover all the
cases in the Recta table. Originally we had a more complicated
proof of the above results: the proof given here was suggested
as a result of reading [Bruins 19521.

REFERENCES

Bruins , E M 1952 Ancient Egyptian Arithmetic Kon. Nederland

Akademie van Wetenschappen A 55(2), 81-91
Gillings, R J 1972 Mathematics in the Time of the Pharaohs
Cambridge, Mass. (MIT Press) 297 p

****xx******

HISTORY OF MATHEMATICS AT THE WHEATON

COLLEGE CONFERENCE "A CHRISTIAN PERSPECTIVE ON THE
FOUNDATIONS OF MATHEMATICS" 28-30 APRIL 1977

The program included an address by Robert Brabenec

(Wheaton College), “The historical shaping of the foundations
of mathematics” and sessions on “Twentieth Century Issues”,
“The Nature of Mathematical Truth”, “The Existence of Mathema-
tical Objects”, “The Concept of the Infinite”, and “Applications”
Ito pedagogy). A proceedings will be published, and another
conference is planned for May 1979.